Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

AIRCHITECT v2: Learning the Hardware Accelerator Design Space through Unified Representations

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read AIrchitect v2 claims that optimal accelerator configurations can be predicted in one pass by a contrastively trained transformer with unified ordinal outputs, reaching 91.17% accuracy on a 100K-sample MAESTRO dataset.

desk verdict A solid ML-for-DSE engineering contribution that learns to mimic ConfuciuX, but the paper never compares against ConfuciuX itself, so the 'replace iterative search' claim is not yet tested. read the letter →

arxiv 2501.09954 v1 pith:4WUJUFAI submitted 2025-01-17 cs.LG cs.AIcs.AR

Jamin Seo , Akshat Ramachandran , Yu-Chuan Chuang , Anirudh Itagi , Tushar Krishna This is my paper · ORCID
classification cs.LGcs.AIcs.AR
keywords designspaceexplorationDNNacceleratortransformercontrastivelearningunifiedordinalvectorsone-shotpredictionMAESTROhardwareresourceallocation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

AIrchitect v2 aims to turn hardware design space exploration (DSE) for DNN accelerators from an iterative, search-based process into a single learned inference. The paper claims that a transformer encoder-decoder, trained with contrastive learning and a new output representation called Unified Ordinal Vectors, can take a workload description and directly predict the optimal number of processing elements and buffer size. On a 100,000-sample dataset generated for a MAESTRO-modeled accelerator, the model reaches 91.17% accuracy in matching the optimal configuration, about 15% better than prior learning-based methods. The same model, applied to DNN and LLM workloads not seen in training, yields hardware choices with roughly 1.7x lower simulated inference latency than the baselines. A sympathetic reader would care because this suggests the expensive search phase of accelerator design can be replaced by a constant-time predictor.

What carries the argument

The load-bearing machinery is the Unified Ordinal Vector (UOV), an output encoding that represents a hardware configuration as a vector of bucket probabilities: the design space is discretized into K ordered buckets, and a value in bucket r_n is encoded with monotonically increasing entries up to that bucket and zeros after it, so the representation carries both which bucket the design falls in (classification) and where inside the bucket it lies (regression). The encoder is trained with a contrastive loss plus a latency-prediction loss to produce a uniform, smooth embedding space resistant to the dataset's long-tailed label distribution, and the decoder is trained with a unification loss to turn embedding points into UOV predictions. This combination is what lets the model make fine-grained one-shot predictions over a design space of roughly $10^{9}$ configurations.

What would settle it

Sample a set of GEMM workloads, run AIrchitect v2, then run an exhaustive or large Monte Carlo search over the full MAESTRO resource configuration space for the same workloads; if an alternative configuration reliably achieves lower simulated latency than the predicted one, the claim that the model identifies optimal design points is not supported.

Watch

Extended reading notes

Core claim

The paper's central claim is that the hardware accelerator design space can be learned end-to-end well enough that optimal resource allocation becomes a one-shot prediction problem. For each layer of a workload, the model receives GEMM tensor dimensions (M, N, K) and a dataflow choice, and outputs the recommended number of processing elements and L2 buffer size. The authors report that this approach achieves 91.17% prediction accuracy on the MAESTRO-based dataset, outperforming AIRCHITECT v1 (77.60%) and GANDSE (84.39%), and that the recommended hardware gives a 1.7x improvement in simulated inference latency on unseen DNN and LLM workloads relative to the baseline techniques. The contribution is presented as a combination of three components: contrastive pretraining that smooths the non-uniform and long-tailed performance landscape, an encoder-decoder transformer that maps inputs into a uniform embedding space, and Unified Ordinal Vectors that blend classification and regression so the model can predict fine-grained configurations without an unmanageable output space.

Load-bearing premise

The dataset labels are generated by ConfuciuX, a reinforcement-learning search, and are treated as the true optimal hardware configurations; if that search often returns locally optimal designs, the reported accuracy measures agreement with those labels rather than genuine design quality.

Editorial extensions

If this is right

  • Hardware resource allocation for a given workload and mapping can be done in a single forward pass, removing the need for iterative sampling during design space exploration.
  • Contrastive pretraining makes the learned embedding robust to non-uniform, non-convex performance landscapes and long-tailed data, which are the properties that hurt earlier supervised DSE models.
  • Because UOV buckets scale sublinearly relative to classification labels, the approach can extend to larger accelerator design spaces without proportional growth in model size.
  • The recommended configurations transfer to unseen models, including LLMs, with about 1.7x lower simulated latency than the compared baselines.
  • UOV is not tied to AIrchitect v2; the paper shows it also improves accuracy and reduces model size when applied to AIrchitect v1.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper's evaluation measures accuracy against ConfuciuX-generated labels, so the 91.17% number is best read as fidelity to that search procedure; whether it equals true optimality is not established by the reported experiments.
  • If contrastive embeddings generalize the way the paper suggests, the same encoder-decoder recipe could apply to other discrete-continuous co-design problems, such as mapping selection, memory hierarchy sizing, or compiler autotuning, though the paper only demonstrates the MAESTRO resource-allocation task.
  • The 1.7x latency gain on unseen LLMs is computed with the same simulator used to generate training labels; an independent cost model or real silicon measurement would be a stronger test of practical value.
  • The deployment pipeline selects one per-layer recommended configuration for a whole model; a natural extension would be to learn a joint model-level recommendation instead of aggregating per-layer choices, which the paper leaves open.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes AIRCHITECT v2, an encoder-decoder transformer for one-shot hardware accelerator design space exploration. The method trains an encoder with a contrastive loss plus a latency-prediction loss, then trains a decoder with UOV heads that combine classification (bucket selection) and regression (within-bucket prediction). The evaluation uses a 100K-sample MAESTRO-based dataset whose labels are generated by running the ConfuciuX RL search, and reports 91.17% layer-level prediction accuracy, outperforming AIRCHITECT v1 (77.60%) and GANDSE (84.39%), plus a model-level deployment study claiming 1.7x inference-latency improvement on unseen DNN/LLM workloads. The paper also releases code, data, and trained models.

Significance. If the empirical claims hold, the paper offers a useful step toward constant-time DSE for accelerator resource allocation, and the public release of the dataset and code is a concrete contribution. The ablations for the contrastive objective, the UOV-versus-classification comparison, and the embedding-space search comparison against VAE are informative. However, the current evaluation does not yet establish the central claim that one-shot inference can replace iterative search: the accuracy numbers are agreement with ConfuciuX-generated labels, and the deployment comparison omits ConfuciuX itself. The reported gains therefore need additional evidence before the paper can be accepted as a demonstration of replacing search-based DSE.

major comments (4)
  1. [Section III-A, Table III] The reported 91.17% accuracy is agreement with labels produced by executing ConfuciuX with latency as the reward, but the paper does not define the accuracy metric (exact match on both PE and buffer size, or per-dimension agreement) and does not establish that ConfuciuX returns globally optimal configurations. Because the labels are generated by a search method, every accuracy number is bounded above by the label generator's search quality; without an oracle or an exhaustive check over the 768-configuration output space, the claim that the model identifies "optimal design points" is not supported. Please define the metric and report label-quality statistics, for example the distribution of latency gaps between ConfuciuX labels and the best configuration found by exhaustive enumeration.
  2. [Section IV-C, Figure 7] The model-level deployment evaluation compares AIRCHITECT v2 with AIRCHITECT v1, GANDSE, and VAESA+BO, but omits ConfuciuX itself, the RL search method whose labels define the task and whose role the paper claims to replace. The 1.7x latency improvement is therefore relative to other learned baselines only; a direct head-to-head with ConfuciuX (with matched number of MAESTRO evaluations or wall-clock time), or with exhaustive enumeration, is required to support the central claim that one-shot inference replaces iterative search.
  3. [Section IV-A, Section IV-D, Figure 8(b)] Hyperparameter K (the number of UOV buckets) is chosen using accuracy and model-size curves in Figure 8(b), but the paper describes only an 80K/20K train/test split with no validation set. If the accuracy curve is computed on the same test split used for Table III, selecting K=16 on the test set leaks test information into the reported 91.17%; please report a validation-based selection procedure and state clearly which split each ablation uses.
  4. [Tables II-III, Figure 7] All accuracy and latency results are single point estimates with no error bars, confidence intervals, or significance tests. Given that training involves stochastic contrastive sampling and 500/100 epochs, the claim of a 15% improvement over baselines needs repeated-seed results or a paired significance test to establish that the gap is not within run-to-run variance.
minor comments (5)
  1. [Section III-D, reference [18]] Reference [18], cited for UOV, is an indoor depth-completion paper (NTrans-Net) and appears unrelated to unified ordinal vectors; please cite the actual source or add a derivation/definition to make the provenance clear.
  2. [Equation (1)] The infoNCE expression in Equation (1) repeats the same exponential terms in the numerator and denominator; please use explicit positive and negative index sets to avoid the apparent redundancy.
  3. [Table II] The first row of Table II is described in the text as training without both objectives and with an L2 loss term, but the table has no row label for that configuration; please make the ablation conditions explicit in the table.
  4. [Figure 7] Figure 7 is normalized to AIRCHITECT v2 with no absolute latency values or explicit workload names in the caption; report actual numbers so readers can assess the claimed 1.7x improvement.
  5. [Abstract and Section VI] The abstract and conclusion use "foundation models" while the body focuses on LLMs; align the terminology, and fix the typo "MAESRO" in Section III-E.

Circularity Check

0 steps flagged · score 2.0 of 10

No derivation reduces to its own inputs; the only circularity-adjacent signal is a minor same-author citation of UOV [18] plus empirically tuned hyperparameters, while the central supervised learning pipeline is self-contained.

full rationale

The central derivation is self-contained supervised learning rather than a circular reduction. The dataset (Section III-A) is generated by executing ConfuciuX [12] with latency as the reward, and the 91.17% accuracy in Table III is agreement with those labels on a held-out split; that makes the evaluation fidelity to an external search method (a benchmark-validity caveat), not a fitted parameter renamed as a prediction. The stage-1 contrastive objective (Eq. 1), stage-2 UOV prediction (Eq. 2), and unification loss (Eq. 3) are each explicitly defined in the paper, with UOV fully specified in Algorithm 1, so the output is not equivalent to the input by construction. The only circularity-adjacent signal is a minor self-citation: Unified Ordinal Vectors is credited to [18], whose authors overlap with the present paper (Akshat Ramachandran), but the representation is specified locally and its effect is measured through ablations; the citation is not load-bearing. Hyperparameters tau=0.4, alpha=0.75, gamma=1, and K=16 are empirically selected on the benchmark, which is standard tuning rather than a definitional shortcut. The deployment comparison omits ConfuciuX itself, so the 1.7x claim is not a test against the label-generating search method; this is a missing-baseline concern, not circularity. Overall, no step in the derivation chain reduces to its own inputs.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new physical entities. The load-bearing assumptions are that MAESTRO and ConfuciuX provide trustworthy ground truth, that the fixed feature encoding captures the design problem, and that UOV bucket membership is a meaningful similarity signal. The main free parameters are loss weights, the contrastive temperature, and the number of buckets, all tuned on the benchmark.

free parameters (4)
  • contrastive temperature tau = 0.4
    Empirically determined for the infoNCE contrastive loss in Section III-C.
  • unification loss weight alpha = 0.75
    Empirically determined weighting between positive and negative UOV terms in Section III-D.
  • unification loss exponent gamma = 1
    Empirically determined exponent in the unification loss in Section III-D.
  • number of UOV buckets K = 16
    Selected via ablation on accuracy and model size in Section IV-D; used for all reported results.
assumptions (4)
  • domain assumption MAESTRO simulator latency is a valid proxy for true hardware performance.
    All ground truth and deployment evaluations use MAESTRO cycle estimates, not measurements on fabricated hardware. Invoked throughout Section IV.
  • domain assumption ConfuciuX-generated labels are the optimal hardware configurations.
    The dataset is generated by running ConfuciuX RL search, and its returned configuration is treated as ground truth for training and accuracy in Section III-A.
  • domain assumption Membership in the same UOV bucket defines semantic similarity for contrastive learning.
    Positive and negative samples are defined by whether they fall into the same or different UOV buckets in Section III-C, which assumes the bucket granularity captures meaningful design similarity.
  • domain assumption The input feature encoding from ConfuciuX (M, N, K, dataflow) is sufficient to determine the optimal configuration.
    The problem formulation in Section III-A fixes L1 buffer size and represents each layer by four features, excluding other architectural parameters that could affect the optimum.

how reviews work

0 comments
Cite this review

Pith. "Pith review of AIRCHITECT v2: Learning the Hardware Accelerator Design Space through Unified Representations." pith.science (2026). https://pith.science/paper/4WUJUFAI

@misc{pith2026250109954,
  author       = {Pith},
  title        = {Pith review of: AIRCHITECT v2: Learning the Hardware Accelerator Design Space through Unified Representations},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/4WUJUFAI}},
  note         = {Machine review of arXiv:2501.09954}
}
read the original abstract

Design space exploration (DSE) plays a crucial role in enabling custom hardware architectures, particularly for emerging applications like AI, where optimized and specialized designs are essential. With the growing complexity of deep neural networks (DNNs) and the introduction of advanced foundational models (FMs), the design space for DNN accelerators is expanding at an exponential rate. Additionally, this space is highly non-uniform and non-convex, making it increasingly difficult to navigate and optimize. Traditional DSE techniques rely on search-based methods, which involve iterative sampling of the design space to find the optimal solution. However, this process is both time-consuming and often fails to converge to the global optima for such design spaces. Recently, AIrchitect v1, the first attempt to address the limitations of search-based techniques, transformed DSE into a constant-time classification problem using recommendation networks. In this work, we propose AIrchitect v2, a more accurate and generalizable learning-based DSE technique applicable to large-scale design spaces that overcomes the shortcomings of earlier approaches. Specifically, we devise an encoder-decoder transformer model that (a) encodes the complex design space into a uniform intermediate representation using contrastive learning and (b) leverages a novel unified representation blending the advantages of classification and regression to effectively explore the large DSE space without sacrificing accuracy. Experimental results evaluated on 10^5 real DNN workloads demonstrate that, on average, AIrchitect v2 outperforms existing techniques by 15% in identifying optimal design points. Furthermore, to demonstrate the generalizability of our method, we evaluate performance on unseen model workloads (LLMs) and attain a 1.7x improvement in inference latency on the identified hardware architecture.

Figures

Figures reproduced from arXiv: 2501.09954 by the authors.

Figure 1
Figure 1. Different design choices yield a wide range of per [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of AIRCHITECT V2, highlighting (1) multi￾head self-attention-based encoder and decoder structure, (2) la￾tent embedding space improved by contrastive learning (3) UOV output representation combining classification and regression. its benefits in smoothening the loss landscape. The core idea of contrastive learning, which is based on the infoNCE loss [25] is, to balance the learning of a data sample (anchor)… view at source ↗
Figure 3
Figure 3. Prominent challenges on DSE dataset: (a) non-uniform [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Complexity of the problem space from Table I, visu [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 2
Figure 2. Figure 2: To effectively learn the complex DSE space, we de [PITH_FULL_IMAGE:figures/full_fig_p003_2.png]
Figure 5
Figure 5. Figure 5: Visualization of embedding space (a) without con [PITH_FULL_IMAGE:figures/full_fig_p004_5.png]
Figure 6
Figure 6. Figure 6: Visualization of how the same configuration is repre [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 9
Figure 9. Figure 9: Effect of UOV on prediction accuracy(red) and model size(green), for AIRCHITECT V1 and AIRCHITECT V2. fine-grained prediction. Additionally, UOV significantly reduces model size, which highlights its applicability in larger design spaces. By demonstrating the advantage…
Figure 8
Figure 8. Figure 8: (a) DSE performance comparison of searching using BO [PITH_FULL_IMAGE:figures/full_fig_p006_8.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. DiffAxE: Diffusion-driven Hardware Accelerator Generation and Design Space Exploration

    cs.AR 2025-08 conditional novelty 6.0 of 10

    DiffAxE uses conditional diffusion models to generate hardware accelerator designs directly from target performance, achieving orders-of-magnitude faster design space exploration with lower error than existing optimiz...

Reference graph

Works this paper leans on

35 extracted references · 22 canonical work pages · cited by 1 Pith paper

  1. [1]

    Algorithm-Hardware Co-Design of Distribution-Aware Logarithmic-Posit Encodings for Efficient DNN Inference

    A. Ramachandran, Z. Wan, G. Jeong, J. Gustafson, and T. Krishna, “Algorithm-hardware co-design of distribution-aware logarithmic-posit encodings for efficient dnn inference,” arXiv preprint arXiv:2403.05465 , 2024

    work page Pith review arXiv 2024

  2. [2]

    Maeri: Enabling flexible dataflow mapping over dnn accelerators via reconfigurable interconnects,

    H. Kwon, A. Samajdar, and T. Krishna, “Maeri: Enabling flexible dataflow mapping over dnn accelerators via reconfigurable interconnects,” ACM SIGPLAN Notices , vol. 53, no. 2, pp. 461–475, 2018

    work page 2018

  3. [3]

    MicroScopiQ: Accelerating Foundational Models through Outlier-Aware Microscaling Quantization

    A. Ramachandran, S. Kundu, and T. Krishna, “Microscopiq: Accelerating foundational models through outlier-aware microscaling quantization,” arXiv preprint arXiv:2411.05282 , 2024

    work page Pith review arXiv 2024

  4. [4]

    Dosa: Differentiable model-based one-loop search for dnn accelerators,

    C. Hong, Q. Huang, G. Dinh, M. Subedar, and Y . S. Shao, “Dosa: Differentiable model-based one-loop search for dnn accelerators,” in Proceedings of the 56th Annual IEEE/ACM International Symposium on Microarchitecture, ser. MICRO ’23. New York, NY , USA: Association for Computing Machinery, 2023, p. 209–224. [Online]. Available: https://doi.org/10.1145/36...

    arXiv 2023

  5. [5]

    Airchitect: Learning custom architecture design and mapping space,

    A. Samajdar, J. M. Joseph, M. Denton, and T. Krishna, “Airchitect: Learning custom architecture design and mapping space,” 2021. [Online]. Available: https://arxiv.org/abs/2108.08295

    arXiv 2021

  6. [6]

    Nvdla deep learning accelerator,

    NVIDIA, “Nvdla deep learning accelerator,” http://nvdla.org, 2017

    work page 2017

  7. [7]

    Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,

    Y .-H. Chen, T. Krishna, J. S. Emer, and V . Sze, “Eyeriss: An energy- efficient reconfigurable accelerator for deep convolutional neural net- works,” IEEE Journal of Solid-State Circuits, vol. 52, no. 1, pp. 127–138, 2016

    work page 2016

  8. [8]

    Shidiannao: Shifting vision processing closer to the sensor,

    Z. Du, R. Fasthuber, T. Chen, P. Ienne, L. Li, T. Luo, X. Feng, Y . Chen, and O. Temam, “Shidiannao: Shifting vision processing closer to the sensor,” in International Symposium on Computer Architecture (ISCA) , 2015

    work page 2015

Show all 35 references
  1. [9]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in 2009 IEEE conference on computer vision and pattern recognition . Ieee, 2009, pp. 248–255

    2009

  2. [10]

    Unified perceptual parsing for scene understanding,

    T. Xiao, Y . Liu, B. Zhou, Y . Jiang, and J. Sun, “Unified perceptual parsing for scene understanding,” in Proceedings of the European conference on computer vision (ECCV) , 2018, pp. 418–434

    2018

  3. [11]

    Learn- ing a continuous and reconstructible latent space for hardware accelerator design,

    Q. Huang, C. Hong, J. Wawrzynek, M. Subedar, and Y . S. Shao, “Learn- ing a continuous and reconstructible latent space for hardware accelerator design,” in 2022 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) , 2022, pp. 277–287

    2022

  4. [12]

    Confuciux: Autonomous hardware resource assignment for dnn accelerators using reinforcement learning,

    S.-C. Kao, G. Jeong, and T. Krishna, “Confuciux: Autonomous hardware resource assignment for dnn accelerators using reinforcement learning,” in 2020 53rd Annual IEEE/ACM International Symposium on Microar- chitecture (MICRO), 2020, pp. 622–636

    2020

  5. [13]

    Gamma: Automating the hw mapping of dnn models on accelerators via genetic algorithm,

    S.-C. Kao and T. Krishna, “Gamma: Automating the hw mapping of dnn models on accelerators via genetic algorithm,” in 2020 IEEE/ACM International Conference On Computer Aided Design (ICCAD) , 2020, pp. 1–9

    2020

  6. [14]

    Digamma: Domain- aware genetic algorithm for hw-mapping co-optimization for dnn accel- erators,

    S.-C. Kao, M. Pellauer, A. Parashar, and T. Krishna, “Digamma: Domain- aware genetic algorithm for hw-mapping co-optimization for dnn accel- erators,” in 2022 Design, Automation & Test in Europe Conference & Exhibition (DATE), 2022, pp. 232–237

    2022

  7. [15]

    Hasco: Towards agile hardware and software co-design for tensor computation,

    Q. Xiao, S. Zheng, B. Wu, P. Xu, X. Qian, and Y . Liang, “Hasco: Towards agile hardware and software co-design for tensor computation,” 2021 ACM/IEEE 48th Annual International Symposium on Computer Architecture (ISCA), pp. 1055–1068, 2021

    2021

  8. [16]

    Gandse: Generative adversarial network-based design space exploration for neural network accelerator design,

    L. Feng, W. Liu, C. Guo, K. Tang, C. Zhuo, and Z. Wang, “Gandse: Generative adversarial network-based design space exploration for neural network accelerator design,” ACM Transactions on Design Automation of Electronic Systems, vol. 28, no. 3, pp. 1–20, 2023

    2023

  9. [17]

    A systematic methodology for characterizing scalability of dnn accelerators using scale-sim,

    A. Samajdar, J. M. Joseph, Y . Zhu, P. Whatmough, M. Mattina, and T. Krishna, “A systematic methodology for characterizing scalability of dnn accelerators using scale-sim,” in 2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS) , 2020, pp. 58–68

    2020

  10. [18]

    Ntrans- net: A multi-scale neutrosophic-uncertainty guided transformer network for indoor depth completion,

    A. Ramachandran, A. Dhiman, B. S. Vandrotti, and J. Kim, “Ntrans- net: A multi-scale neutrosophic-uncertainty guided transformer network for indoor depth completion,” in 2023 IEEE International Conference on Image Processing (ICIP) . IEEE, 2023, pp. 905–909

    2023

  11. [19]

    MAESTRO: A data-centric approach to understand reuse, performance, and hardware cost of DNN mappings,

    H. Kwon, P. Chatarasi, V . Sarkar, T. Krishna, M. Pellauer, and A. Parashar, “MAESTRO: A data-centric approach to understand reuse, performance, and hardware cost of DNN mappings,” IEEE Micro, vol. 40, no. 3, pp. 20–29, 2020

    2020

  12. [20]

    A systematic methodology for characterizing scalability of dnn accelerators using scale-sim,

    A. Samajdar, J. M. Joseph, Y . Zhu, P. Whatmough, M. Mattina, and T. Krishna, “A systematic methodology for characterizing scalability of dnn accelerators using scale-sim,” in 2020 IEEE International Symposium on Performance Analysis of Systems and Software (ISPASS). IEEE, 202...

    2020

  13. [21]

    Contrastive quant: quantization makes stronger contrastive learning,

    Y . Fu, Q. Yu, M. Li, X. Ouyang, V . Chandra, and Y . Lin, “Contrastive quant: quantization makes stronger contrastive learning,” in Proceedings of the 59th ACM/IEEE Design Automation Conference , 2022, pp. 205– 210

    2022

  14. [22]

    Synergistic self- supervised and quantization learning,

    Y .-H. Cao, P. Sun, Y . Huang, J. Wu, and S. Zhou, “Synergistic self- supervised and quantization learning,” in European Conference on Com- puter Vision. Springer, 2022, pp. 587–604

    2022

  15. [23]

    Jumping through local minima: Quantization in the loss landscape of vision transformers,

    N. Frumkin, D. Gope, and D. Marculescu, “Jumping through local minima: Quantization in the loss landscape of vision transformers,” in Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023, pp. 16 978–16 988

    2023

  16. [24]

    Clamp-vit: Contrastive data-free learning for adaptive post-training quantization of vits,

    A. Ramachandran, S. Kundu, and T. Krishna, “Clamp-vit: Contrastive data-free learning for adaptive post-training quantization of vits,” arXiv preprint arXiv:2407.05266, 2024

    2024 arXiv

  17. [25]

    Positive–negative equal contrastive loss for semantic segmentation,

    J. Wang, J. Li, W. Li, L. Xuan, T. Zhang, and W. Wang, “Positive–negative equal contrastive loss for semantic segmentation,” Neurocomputing, vol. 535, pp. 13–24, 2023

    2023

  18. [26]

    Targeted supervised contrastive learning for long-tailed recognition,

    T. Li, P. Cao, Y . Yuan, L. Fan, Y . Yang, R. S. Feris, P. Indyk, and D. Katabi, “Targeted supervised contrastive learning for long-tailed recognition,” in Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , 2022, pp. 6918–6928

    2022

  19. [27]

    A meta-analysis of overfitting in machine learning,

    R. Roelofs, V . Shankar, B. Recht, S. Fridovich-Keil, M. Hardt, J. Miller, and L. Schmidt, “A meta-analysis of overfitting in machine learning,” Advances in Neural Information Processing Systems , vol. 32, 2019

    2019

  20. [28]

    Attention is all you need,

    A. Vaswani, “Attention is all you need,” Advances in Neural Information Processing Systems, 2017

    2017

  21. [29]

    Automatic chemical design using a data-driven continuous representation of molecules,

    R. G ´omez-Bombarelli, J. N. Wei, D. Duvenaud, J. M. Hern ´andez-Lobato, B. S´anchez-Lengeling, D. Sheberla, J. Aguilera-Iparraguirre, T. D. Hirzel, R. P. Adams, and A. Aspuru-Guzik, “Automatic chemical design using a data-driven continuous representation of molecules,” ACS ce...

    2018

  22. [30]

    Occdepth: A depth-aware method for 3d semantic scene completion,

    R. Miao, W. Liu, M. Chen, Z. Gong, W. Xu, C. Hu, and S. Zhou, “Occdepth: A depth-aware method for 3d semantic scene completion,” arXiv preprint arXiv:2302.13540 , 2023

    2023 arXiv

  23. [31]

    Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection,

    X. Li, W. Wang, L. Wu, S. Chen, X. Hu, J. Li, J. Tang, and J. Yang, “Generalized focal loss: Learning qualified and distributed bounding boxes for dense object detection,” Advances in Neural Information Processing Systems, vol. 33, pp. 21 002–21 012, 2020

    2020

  24. [32]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE conference on computer vision and pattern recognition , 2016, pp. 770–778

    2016

  25. [33]

    Llama 2: Open foundation and fine-tuned chat models,

    H. Touvron, L. Martin, K. Stone, P. Albert, A. Almahairi, Y . Babaei, N. Bashlykov, S. Batra, P. Bhargava, S. Bhosale et al., “Llama 2: Open foundation and fine-tuned chat models,”arXiv preprint arXiv:2307.09288, 2023

    2023 arXiv

  26. [34]

    The llama 3 herd of models,

    A. Dubey, A. Jauhri, A. Pandey, A. Kadian, A. Al-Dahle, A. Letman, A. Mathur, A. Schelten, A. Yang, A. Fan et al. , “The llama 3 herd of models,” arXiv preprint arXiv:2407.21783 , 2024

    2024 arXiv

  27. [35]

    Mind mappings: Enabling efficient algorithm-accelerator mapping space search,

    K. Hegde, P.-A. Tsai, S. Huang, V . Chandra, A. Parashar, and C. W. Fletcher, “Mind mappings: Enabling efficient algorithm-accelerator mapping space search,” in Proceedings of the 26th ACM International Conference on Architectural Support for Programming Languages and Operatin...

    2021

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.